Codex Desktop becomes unresponsive while Tailscale reports no internet access; Tailscale recovers after Codex exits

What version of the Codex App are you using (From “About Codex” dialog)?

26.908.4834.0

What subscription do you have?

ChatGPT Pro

What platform is your computer?

Microsoft Windows NT 10.0.26200.0 x64 (Windows 11 Pro 25H2, OS build 26200.9445)

What issue are you seeing?

On September 13, 2026, approximately 1:40–2:10 PM EDT, Codex stopped accepting or transmitting new messages during a long-running task. At approximately the same time, the Tailscale client reported “no internet access,” although normal web browsing continued to work. A local mobile companion service reachable through Tailscale became unavailable.

Fully closing Codex immediately restored Tailscale connectivity. Restarting Codex restored Codex functionality.

The task appeared to remain active and displayed a status approximately referring to “working on” or “coordinating” independent threads. This is not a verbatim quote; the exact wording was not preserved and no screenshot was captured.

Impact: Codex work and new messages were unavailable until restart, and the mobile companion could not reach its PC-hosted service through Tailscale.

What steps can reproduce the bug?

This has occurred once and is not yet reliably reproducible. The observed sequence was:

1. Codex Desktop was running a long-running task in a trusted project with unrestricted filesystem access and no active filesystem sandbox.
2. After some time, Codex stopped accepting or transmitting new messages.
3. At approximately the same time, Tailscale reported “no internet access,” although ordinary web browsing continued working.
4. A local mobile companion service reachable through Tailscale became unavailable.
5. Codex Desktop was fully closed.
6. Tailscale connectivity returned immediately after Codex exited.
7. Codex functioned normally again after relaunch.

No session ID, exact status text, screenshot, or contemporaneous token/context-window reading was preserved.

What is the expected behavior?

Codex should remain responsive during long-running and multi-threaded work and should not coincide with loss of VPN or virtual-network connectivity.

Additional information

Environment and diagnostics:

• Incident time: September 13, 2026, approximately 1:40–2:10 PM EDT
• Tailscale for Windows: 1.102.4
• The affected task was a trusted project with unrestricted local filesystem access and no active filesystem sandbox.
• Normal web browsing continued to work during the incident.
• No relevant Windows Application or System errors were found afterward.
• No evidence of general RAM or GPU exhaustion was found.
• No associated application crash or mobile-client ANR was found.

Workload context:

At the time of the incident, Codex was handling a long-running portfolio-management and project-coordination task involving multiple independent project workers/threads. The work included inspecting project state and coordinating separate project-specific tasks. It was not running GPU-intensive generation, a large compilation, or another known system-heavy process when the failure was noticed.

Possible area to investigate:

There may have been an interaction involving Codex's local tool host, networking resources, persistent connections, or Windows virtual-network handling. Causality is unknown, but Tailscale's immediate recovery after Codex exited suggests a meaningful correlation.

If the issue recurs, I can capture tailscale status, Codex process and network state, timestamps, and screenshots before terminating Codex.

No source files or tests are identified. Start by reproducing a long-running Codex Desktop task on Windows while collecting Tailscale status, Codex process and network state, timestamps, and screenshots. Done means Codex remains responsive and Tailscale connectivity does not disappear during comparable work.